🚀 go-pugleaf

           Department of Computer Science, University of Toronto
         (SF = Sandford Fleming Building, 10 King's College Road)

       -------------------------------------------------------------

                                 DATABASE
              SF1105, at 2:00 p.m., Thursday 19 November 1992

                             Chaitanya K. Baru
                             IBM Toronto Labs

       "Data Placement in Shared-Nothing Parallel Database Systems"

In this talk, we will address the issue of data placement in multicomputer
database systems and discuss the optimization of the placement of database
tables across the nodes of a scalable, shared-nothing parallel computer
system.  Two important issues in data placement are (i) finding the
appropriate number of nodes across which a given database table is to be
partitioned and (ii) assigning the table partitions to specific nodes in
the system.  Past work has not addressed all aspects of the declustering
and assignment problems.  The data placement problem is formulated as an
optimization problem using the notion of a database placement state.  We
present several algorithms that employ this formulation to solve the
declustering and assignment problems in data placement.  Experimental
results using different complex query and mixed workloads for a given
database, are also presented.  The results indicate that data placement can
critically affect the performance--both, response time as well as
throughput--of a parallel database system.